Innodata (INOD) Q3 2026 Earnings Call Transcript

Innodata reported Q3 2026 revenue of $95.6 million, up 53% year over year and 15% above consensus; adjusted EBITDA was $17.7 million versus $15.4 million expected, and diluted EPS of $0.29 beat the $0.17 consensus. The company raised its 2026 revenue-growth outlook to 42% or more from 40%–42%, citing new AI-related business, while expecting a $12 million Q4 revenue headwind after one program with its traditionally largest customer was ended. Management cited $29 million in new program value with its now-largest customer and forecast more than $20 million of 2027 revenue from another hyperscaler customer, versus $3.1 million in trailing-12-month revenue.

Analysis

INOD’s key debate is whether new programs can replace revenue quickly enough to offset the Q4 gap—not whether demand for AI data is growing. Management’s expected Q1 replacement is not yet secured: it described late-stage programs, not a confirmed award, and explicitly said labor skills are not always reusable. That creates asymmetric execution risk: idle or retrained labor could pressure margins even if revenue later recovers. Customer diversification is directionally positive, but the change in which customer is largest does not itself establish lower concentration; verify customer-level revenue shares in filings.

The reported EPS beat is less durable than the operating result suggests: the year-to-date tax rate is well below management’s stated long-term range, while gross margin remains mix-sensitive at its 40% target. Treat the $29m contract value and prospective 2027 customer revenues as pipeline indicators, not booked revenue. Physical-AI data, agentic evaluation and enterprise implementation could extend the business beyond episodic labeling, but recurring economics and customer renewal behavior are unproven. A pilot that fails to scale, delayed federal contracting, or a shift in model-builder budgets could unwind the growth premium. Partnerships involving Anthropic, Snowflake and Google Cloud may open channels, but the call provides no quantified partner-sourced revenue.

Near term, the Q4 customer-program loss and confirmation of replacement awards dominate. Over 1–3 months, Q4 guidance and signed-work conversion are the catalysts; over 6–18 months, recurring physical-AI and agentic programs are the structural test. With no valuation or price data, the upside case is operationally credible but not enough to justify chasing a post-beat move.

Key Decisions for Investors

• Keep INOD on a catalyst watch; consider a staged long only after management confirms signed replacement work and the Q4 revenue impact is understood. Do not underwrite the expected Q1 recovery as contracted.
• Use the next results to test revenue quality: track customer concentration, program renewals, utilization/redeployment costs, adjusted gross margin versus the 40% target, and cash generation excluding customer prepayments.
• Falsify the recovery thesis if replacement awards slip beyond Q1, the Q4 gap persists into guidance, or gross margin falls materially below target as new programs ramp. A failed robotics pilot or delayed federal award would further weaken the 2027 growth case.
• Avoid treating Anthropic, Snowflake or Google Cloud partnerships as standalone beneficiaries of this call: partner-sourced sales and economics were not quantified.